Application of GLAD-ARD products for characterization of rapid land transformation in the Delhi Metropolitan Area

Keywords: urban remote sensing, Indian cities, urban spatial research, machine learning, supervised remote sensing, land cover classification
Abstract Type: Virtual Paper Abstract

Abstract

Transformation of land from urbanization in Remote Sensing studies has been conventionally represented by changing urban footprints. Preparing historic time-series classification maps of rapidly changing urban areas presents a challenge due to the limited availability of high-resolution imagery. And, the more widely available medium-resolution optical imagery, such as those from Landsat, is less utilized to characterize urban land uses due to the lack of details. In this study, we utilize the GLAD Landsat Analysis Ready Data (ARD) to examine its capabilities in characterizing the change in the heterogeneous landscape of Delhi’s Metropolitan Area between 2000 and 2020. The 16-day product, pre-processed for spatial and temporal consistency, was used to derive phenological metrics and assemble annual images for every four years. We also gathered training and validation data for modeling through extensive fieldwork, identifying nine classes that describe the vegetation land covers and land uses related to the built environment of the study area. A Random Forest (RF) classification model for 2020 was then prepared and extrapolated to images from previous years. The RF model shows an overall accuracy of 82.45%. Two classes of the built environment, depicting High-density and Low-density settlements, exhibit an accuracy of over 72%, an acceptable threshold for a classification study. The model prediction of land classes extrapolated to other years in the time series indicates the need for external data to improve the accuracy of the time series. The next step would involve investigating the change with the growth of transportation infrastructure and the land market.